threads / bug / 32950

[BUG] Infinite make recursion when configure.ac changes

Subject: [BUG] Infinite make recursion when configure.ac changes

## tl;dr

5 messages between Feb 20, 2013 and Feb 21, 2013.

replies: 4people: 3as markdown or json

Stefano Lattarini· Feb 20, 2013, 08:44 UTC · lore
>From a pristine master checkout:
  $ make configure && ./configure make
  ... # All successfull
  $ touch configure.ac
  $ make
    GEN config.status
  make[1]: Entering directory `/storage/home/stefano/git/src'
    GEN config.status
  make[2]: Entering directory `/storage/home/stefano/git/src'
    GEN config.status
  make[3]: Entering directory `/storage/home/stefano/git/src'
    GEN config.status
  make[4]: Entering directory `/storage/home/stefano/git/src'
    GEN config.status
  make[5]: Entering directory `/storage/home/stefano/git/src'
    GEN config.status
  ...
and I have to hit ^C to interrupt that recursion.

This seems due to the change in commit v1.7.12.4-1-g1226504: the issue is still present there, but no longer is in the preceding commit 7e201053 (a.k.a. v1.7.12.4).

I haven't investigated this any further for the moment.
Regards,
  Stefano
Jeff King· Feb 21, 2013, 06:04 UTC · re: Stefano Lattarini · lore

Re: [BUG] Infinite make recursion when configure.ac changes

On Wed, Feb 20, 2013 at 09:44:58AM +0100, Stefano Lattarini wrote:
Show 20 quoted lines
> From a pristine master checkout:
> 
>   $ make configure && ./configure make
>   ... # All successfull
>   $ touch configure.ac
>   $ make
>     GEN config.status
>   make[1]: Entering directory `/storage/home/stefano/git/src'
>     GEN config.status
>   make[2]: Entering directory `/storage/home/stefano/git/src'
>     GEN config.status
>   make[3]: Entering directory `/storage/home/stefano/git/src'
>     GEN config.status
>   make[4]: Entering directory `/storage/home/stefano/git/src'
>     GEN config.status
>   make[5]: Entering directory `/storage/home/stefano/git/src'
>     GEN config.status
>   ...
> 
> and I have to hit ^C to interrupt that recursion.
I can easily replicate it here.
Show 5 quoted lines
> This seems due to the change in commit v1.7.12.4-1-g1226504: the
> issue is still present there, but no longer is in the preceding
> commit 7e201053 (a.k.a. v1.7.12.4).
> 
> I haven't investigated this any further for the moment.

Hmm. It looks like config.status depends on configure.ac. So make rebuilds config.status, which runs "make configure", which includes "config.mak.autogen", leading "make" to think that it should rebuild the include file to make sure it is up to date. The include file depends on "config.status", which needs to be rebuilt due to configure.ac, which entails running "make configure", and so on.

So the real problem is that make things that "make configure" depends on "config.mak.autogen" being up to date, which isn't true at all; the whole point is to make a new configure script so we can write a new config.mak.autogen. The simplest thing is to just avoid re-running "make" to get the configure script; the only thing we are gaining is that we don't have to repeat the build recipe, but we can sneak around that by sticking it in a variable. Something like this seems to work:

diff --git a/Makefile b/Makefile
index 3b2c92c..ee1c0b0 100644
--- a/Makefile
+++ b/Makefile
@@ -1871,12 +1871,14 @@ configure: configure.ac GIT-VERSION-FILE
 	mv $@+ $@
 endif # NO_PYTHON
 
+CONFIGURE_RECIPE = $(RM) configure configure.ac+ && \
+		   sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
+			configure.ac >configure.ac+ && \
+		   autoconf -o configure configure.ac+ && \
+		   $(RM) configure.ac+
+
 configure: configure.ac GIT-VERSION-FILE
-	$(QUIET_GEN)$(RM) $@ $<+ && \
-	sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
-	    $< > $<+ && \
-	autoconf -o $@ $<+ && \
-	$(RM) $<+
+	$(QUIET_GEN)$(CONFIGURE_RECIPE)
 
 ifdef AUTOCONFIGURED
 # We avoid depending on 'configure' here, because it gets rebuilt
@@ -1885,7 +1887,7 @@ config.status: configure.ac
 # do want to recheck when the platform/environment detection logic
 # changes, hence this depends on configure.ac.
 config.status: configure.ac
-	$(QUIET_GEN)$(MAKE) configure && \
+	$(QUIET_GEN)$(CONFIGURE_RECIPE) && \
 	if test -f config.status; then \
 	  ./config.status --recheck; \
 	else \

-Peff
Junio C Hamano· Feb 21, 2013, 06:10 UTC · re: Jeff King · lore

Re: [BUG] Infinite make recursion when configure.ac changes

Jeff King <peff@peff.net> writes:
Show 14 quoted lines
> I can easily replicate it here.
>
>> This seems due to the change in commit v1.7.12.4-1-g1226504: the
>> issue is still present there, but no longer is in the preceding
>> commit 7e201053 (a.k.a. v1.7.12.4).
>> 
>> I haven't investigated this any further for the moment.
>
> Hmm. It looks like config.status depends on configure.ac. So make
> rebuilds config.status, which runs "make configure", which includes
> "config.mak.autogen", leading "make" to think that it should rebuild the
> include file to make sure it is up to date. The include file depends on
> "config.status", which needs to be rebuilt due to configure.ac, which
> entails running "make configure", and so on.

I noticed this while looking at the other autoconf patch yesterday, but I was otherwise occupied in the evening and did not pursue it further. Thanks for looking into it.

This may be an unrelated issue, but I've always thought that it was strange and extremely unintuitive that running "make configure" once only creates config.mak.autogen, while running it once again after that (i.e. while having config.mak.autogen in the tree) seems to run the resulting "./configure" as well. Maybe it is just me.

Jeff King· Feb 21, 2013, 06:26 UTC · re: Junio C Hamano · lore

Re: [BUG] Infinite make recursion when configure.ac changes

On Wed, Feb 20, 2013 at 10:10:42PM -0800, Junio C Hamano wrote:
> I noticed this while looking at the other autoconf patch yesterday,
> but I was otherwise occupied in the evening and did not pursue it
> further.  Thanks for looking into it.

Here's the patch with a commit message. I'm pretty sure this is the right thing to do. I was tempted to find a way of telling make "no, don't include config.mak.autogen, but that actually _isn't_ right. It might be defining $(RM) or some other variable that we use in the recipe. By including it inline, we will use whatever is in the current config.mak.autogen (that we read when we started this make invocation), which is better than nothing.

Show 5 quoted lines
> This may be an unrelated issue, but I've always thought that it was
> strange and extremely unintuitive that running "make configure" once
> only creates config.mak.autogen, while running it once again after
> that (i.e. while having config.mak.autogen in the tree) seems to run
> the resulting "./configure" as well. Maybe it is just me.

It's not the "configure" target that causes it to run. It's the fact that we include config.mak.autogen, which we then tell make can be rebuilt by running ./configure. _But_ we only tell it so if one already exists (since otherwise everybody would get automake cruft). This is the "ifdef AUTOCONFIGURED" you see.

So I think the rule makes sense. Once you have told the Makefile that you are interested in autoconf (by running configure once), then it rebuilds it as necessary, and it should be consistent. You can opt back out of it be removing config.mak.autogen.

Anyway, here is the patch to fix the loop.
-- >8 --
Subject: [PATCH] Makefile: avoid infinite loop on configure.ac change

If you are using autoconf and change the configure.ac, the Makefile will notice that config.status is older than configure.ac, and will attempt to rebuild and re-run the configure script to pick up your changes. The first step in doing so is to run "make configure". Unfortunately, this tries to include config.mak.autogen, which depends on config.status, which depends on configure.ac; so we must rebuild config.status. Which leads to us running "make configure", and so on.

It's easy to demonstrate with:
  make configure
  ./configure
  touch configure.ac
  make

We can break this cycle by not re-invoking make to build "configure", and instead just putting its rules inline into our config.status rebuild procedure. We can avoid a copy by factoring the rules into a make variable.

Signed-off-by: Jeff King <peff@peff.net>
---
 Makefile | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
diff --git a/Makefile b/Makefile
index 3b2c92c..ee1c0b0 100644
--- a/Makefile
+++ b/Makefile
@@ -1871,12 +1871,14 @@ configure: configure.ac GIT-VERSION-FILE
 	mv $@+ $@
 endif # NO_PYTHON
 
+CONFIGURE_RECIPE = $(RM) configure configure.ac+ && \
+		   sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
+			configure.ac >configure.ac+ && \
+		   autoconf -o configure configure.ac+ && \
+		   $(RM) configure.ac+
+
 configure: configure.ac GIT-VERSION-FILE
-	$(QUIET_GEN)$(RM) $@ $<+ && \
-	sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
-	    $< > $<+ && \
-	autoconf -o $@ $<+ && \
-	$(RM) $<+
+	$(QUIET_GEN)$(CONFIGURE_RECIPE)
 
 ifdef AUTOCONFIGURED
 # We avoid depending on 'configure' here, because it gets rebuilt
@@ -1885,7 +1887,7 @@ config.status: configure.ac
 # do want to recheck when the platform/environment detection logic
 # changes, hence this depends on configure.ac.
 config.status: configure.ac
-	$(QUIET_GEN)$(MAKE) configure && \
+	$(QUIET_GEN)$(CONFIGURE_RECIPE) && \
 	if test -f config.status; then \
 	  ./config.status --recheck; \
 	else \
-- 
1.8.1.4.4.g265d2fa
Junio C Hamano· Feb 21, 2013, 07:59 UTC · re: Jeff King · lore

Re: [BUG] Infinite make recursion when configure.ac changes

Jeff King <peff@peff.net> writes:
Show 29 quoted lines
> Anyway, here is the patch to fix the loop.
>
> -- >8 --
> Subject: [PATCH] Makefile: avoid infinite loop on configure.ac change
>
> If you are using autoconf and change the configure.ac, the
> Makefile will notice that config.status is older than
> configure.ac, and will attempt to rebuild and re-run the
> configure script to pick up your changes. The first step in
> doing so is to run "make configure". Unfortunately, this
> tries to include config.mak.autogen, which depends on
> config.status, which depends on configure.ac; so we must
> rebuild config.status. Which leads to us running "make
> configure", and so on.
>
> It's easy to demonstrate with:
>
>   make configure
>   ./configure
>   touch configure.ac
>   make
>
> We can break this cycle by not re-invoking make to build
> "configure", and instead just putting its rules inline into
> our config.status rebuild procedure.  We can avoid a copy by
> factoring the rules into a make variable.
>
> Signed-off-by: Jeff King <peff@peff.net>
> ---
Thanks.  This needs to go to both v1.8.1.x and v1.8.2.
Will apply.
Show 36 quoted lines
>  Makefile | 14 ++++++++------
>  1 file changed, 8 insertions(+), 6 deletions(-)
>
> diff --git a/Makefile b/Makefile
> index 3b2c92c..ee1c0b0 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -1871,12 +1871,14 @@ configure: configure.ac GIT-VERSION-FILE
>  	mv $@+ $@
>  endif # NO_PYTHON
>  
> +CONFIGURE_RECIPE = $(RM) configure configure.ac+ && \
> +		   sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
> +			configure.ac >configure.ac+ && \
> +		   autoconf -o configure configure.ac+ && \
> +		   $(RM) configure.ac+
> +
>  configure: configure.ac GIT-VERSION-FILE
> -	$(QUIET_GEN)$(RM) $@ $<+ && \
> -	sed -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
> -	    $< > $<+ && \
> -	autoconf -o $@ $<+ && \
> -	$(RM) $<+
> +	$(QUIET_GEN)$(CONFIGURE_RECIPE)
>  
>  ifdef AUTOCONFIGURED
>  # We avoid depending on 'configure' here, because it gets rebuilt
> @@ -1885,7 +1887,7 @@ config.status: configure.ac
>  # do want to recheck when the platform/environment detection logic
>  # changes, hence this depends on configure.ac.
>  config.status: configure.ac
> -	$(QUIET_GEN)$(MAKE) configure && \
> +	$(QUIET_GEN)$(CONFIGURE_RECIPE) && \
>  	if test -f config.status; then \
>  	  ./config.status --recheck; \
>  	else \

← back to recent threads